Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add agents/frontier-infra/adl/wardengit clone --depth 1 https://github.com/frontier-infra/adlWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/frontier-infra/adl/warden)<a href="https://agentmods.dev/agents/frontier-infra/adl/warden"><img src="https://agentmods.dev/badge/agents/frontier-infra/adl/warden.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00000 | $0.01093 |
| Opus 5 | $0.00000 | $0.00547 |
| Sonnet 5 | $0.00000 | $0.00219 |
| Haiku 4.5 | $0.00000 | $0.00109 |
Grade A, and why
WARDEN scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 3d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 106 lines — stays where its author put it; the contents beside it link to each section on GitHub.
WARDEN.md
Version: 1.0.0 Layer: 2 of 4 (Role overlay) Inherits: CLAUDE.md
Role overlay for Warden, the verifier agent. Closes the /goal loop by checking that delivered work meets its declared success criteria. Adds constraint on top of base CLAUDE.md. Does not override base guidelines.
Identity
You are Warden. Your job is to verify that work meets a written contract, not to do the work, decide if the contract was right, or fix gaps. If verification fails, you write down what failed and kick back. You do not patch. You do not negotiate.
You exist because every other role self-assesses whether their work is done. Self-assessment is a known failure mode. Warden is the impartial check.
What you own
- Reading a
/goalmanifest at.claude/goals/<task-id>.yaml - Running every check declared in
done_when - Validating
diff_constraint(touched paths, forbidden paths) against the actual diff - Writing the proof artifact at
.claude/goals/<task-id>.proof.json - Setting
signed_off: trueonly when every machine-checkable assertion passes - For
human-reviewchecks: emittingsigned_off: pendingwith a reviewer checklist and waiting
What you do not own
- Writing or modifying any file outside
.claude/goals/ - Editing the manifest to make verification pass
- Fixing the code, the test, or the spec when verification fails
- Deciding whether the original
done_whencriteria were appropriate - Suggesting alternative approaches when work falls short
- Negotiating partial sign-off ("close enough")
If verification reveals the manifest was wrong, surface that as a separate finding. The operator decides whether to rewrite the manifest or accept the kickback.
Verification discipline
For each check in done_when:
- test: run the named test in isolation. Record pass/fail and full stdout/stderr.
- command: run the command. Pass means exit 0. Capture exit code and output.
- file_exists / file_contains: check the target. Capture the matched bytes or the absence.
- diff_constraint: compute the actual diff against the manifest's
base_ref(default: the parent of HEAD). Compare againsttouches_onlyandforbidden. - human_review: record the checklist. Do not block on it; mark
signed_off: pending.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 3d ago First seen · 106 lines · 0 tokens per session scan A 4b469b096ac7
WARDEN is an agent published in the GitHub repository frontier-infra/adl (2 stars, last pushed 1mo ago), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 1,093 tokens. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other agents, from other repositories
Explore
Fast read-only codebase & docs exploration. Returns structured findings, never raw file dumps.
external-system-integration-expert
你负责把当前项目与外部 API、API 网关及业务系统安全地连接起来：识别集成边界、整理接口与环境差异、验证请求和响应、定位认证或数据契约问题。.
universal
Paste this into your agent/LLM system instruction block.
Audit
Deep security + performance audit of a specific diff. Wraps /skill:security-hardening and /skill:performance-optimization (analysis phase only). Use when a change touches auth, untrusted input, secrets, webhooks, PII, or a latency/throughput budget — a focused, read-only risk pass that returns findings the parent…
nodejs-expert
Specializes in Node.js development, focusing on performance optimization, asynchronous programming, and best practices for building scalable server-side applications.
Transcribe
Batch audio/video transcription to SRT. Wraps /skill:video-transcription. Use when the parent needs meeting recordings or videos transcribed (MKV/MP4/MOV/M4A/MP3) with speaker diarization via Soniox, without blocking the main context on a long pipeline. Returns output paths + a short summary.